Researchers developed a simpler AI model to predict heart failure mortality
The new SIMPLE-HF model uses 11 basic clinical variables to outperform existing tools in identifying high-risk patients.
A new study published in npj Digital Medicine details a streamlined AI tool designed to assist doctors in managing heart failure patients. Researchers created a model called SIMPLE-HF, which stands for Simplified Intelligent Mortality Prediction for Longitudinal EHRs in HF patients.
To build this tool, the team analyzed health records from 373,389 adults. They started with a complex Transformer model, a type of AI architecture skilled at finding patterns in large datasets, and then simplified it to focus on only 11 key factors. These include a patient's age, body mass index, existing health conditions, and current medications.
SIMPLE-HF achieved superior discrimination versus MAGGIC-EHR and was well calibrated.
When tested, the 11-variable model showed better performance than the standard MAGGIC-EHR score. It achieved a C-index of 0.801 for predicting mortality over a 12-month period, compared to 0.735 for the traditional method. The researchers noted that at a specific high-risk threshold, their tool correctly identified 177 true events per 1000 patients, nearly doubling the 89 events caught by the standard score.
This approach helps bridge the gap between heavy, data-hungry AI systems and the realities of daily clinical workflows. By relying on information already collected at the point of care, it avoids the need for specialized tests like echocardiography.
Have you considered how much faster medical decisions could be if doctors had real-time predictive insights during a standard checkup?
Despite these results, the authors emphasize that this is not yet ready for hospital use. The model still requires external validation across different patient groups and prospective testing in real clinical settings before it can be adopted into standard medical practice.
Source: npj Digital Medicine